Feat: Add an Astron MaaS screenshot-to-code skill

[FenjuFu]

Summary

Add a scenario-level iflytek-screenshot-to-code skill that turns a UI screenshot into runnable frontend code through an Astron MaaS vision model.

The workflow is inspired by abi/screenshot-to-code, while keeping this repository's skill self-contained and using iFLYTEK's OpenAI-compatible MaaS endpoint.

Motivation

The repository already exposes atomic OCR and image-understanding capabilities. Screenshot-to-code is a useful next composition: it combines visual understanding, layout reconstruction, and code generation into a reusable agent workflow. It also matches the repository roadmap toward scenario-oriented skills.

Proposed scope

• Add skills/iflytek-screenshot-to-code/ with a focused SKILL.md and usage documentation.
• Add a dependency-free Python CLI that:
  • accepts PNG, JPEG, or WebP screenshots;
  • sends a base64 data URL to the Astron MaaS OpenAI-compatible /chat/completions API;
  • supports a small set of explicit output stacks;
  • writes the generated code atomically to a caller-selected path;
  • reports API and malformed-response errors without exposing credentials.
• Configure the endpoint, model ID, and API key through environment variables or CLI flags.
• Add offline unit tests with a mocked HTTP transport; CI must not require MaaS credentials.
• Document upstream inspiration and avoid vendoring code from screenshot-to-code.

Acceptance criteria

• The skill is discoverable for screenshot/mockup-to-code requests.
• Local validation rejects missing files, unsupported formats, empty model responses, and ambiguous multi-file output.
• The default output is a single self-contained HTML file; framework modes clearly document their single-file contract.
• Tests cover request construction, Markdown-fence normalization, response parsing, and output behavior.